feat(eve): productionize Browserbase extension 0.2

[shrey150]

Summary

• publish the Stagehand-owned @browserbasehq/eve extension from this repository
• expose Stagehand V4 Code Mode as browserbase__run, browserbase__snapshot, and browserbase__screenshot
• bundle the canonical Core compatibility runtime into the published package while keeping Stagehand and the Browserbase SDK as asset-preserving external dependencies
• document mounting the package in an existing Eve agent; no repo-owned example agent or model choice is added
• wire Eve into Changesets, changelog consolidation, Turbo, and the TypeScript release build

This is stacked on #2824, which propagates browser-side browser.close() requests to the host lifecycle owner. Version 0.2 replaces the legacy focused Browserbase tools with the Stagehand V4 Code Mode contract. Eve built-in web_search and web_fetch remain unchanged; #2822 adds optional Browserbase overrides.

Release setup

Before the first publish, configure npm trusted publishing for @browserbasehq/eve with repository browserbase/stagehand and workflow .github/workflows/release.yml. The included minor changeset publishes this as 0.2.0.

After 0.2.0 is available, the duplicate package in browserbase/integrations can be retired and its pending 0.2 release should not be published.

Validation

• Eve unit tests: 20/20
• Eve typecheck, extension build, focused format, and focused lint checks
• packed package passes publint and has no private Core runtime dependency
• clean Eve 0.44.4 consumer build preserves stagehand-extension.zip and exposes exactly the three namespaced Code Mode tools
• real Eve model loop: navigation, snapshot, state across calls, deliberate JavaScript error, healthy recovery, hydrated snapshot action, screenshot, final URL/title, and browser.close() all passed
• a later Eve turn created fresh browser resources and closed them; Browserbase reported both sessions COMPLETED, with zero sessions left running

[changeset-bot]

🦋 Changeset detected

Latest commit: 87a393b

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@browserbasehq/eve	Minor

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[shrey150]

any reason this was changed to 5.4 mini? would prefer 5.6 luna

[shrey150]

Addressed in aacd83be9 — restored the example default to gpt-5.6-luna while retaining the environment override.

[shrey150]

Follow-up: the private example workspace has now been removed entirely in 8441da9, so this package no longer owns an Eve agent or model default.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	npm/​eve@​0.44.4
	npm/​ai@​7.0.79

View full report

[cubic-dev-ai]

2 issues found and verified against the latest diff

Confidence score: 4/5

• In scripts/release/consolidate-changelogs.ts, Eve is now included in packageChangelogs, but the release tests do not cover its package path or label, so Eve-generated changelogs could be silently omitted or mishandled—add a focused consolidation test.
• In packages/integrations/eve/extension/tools/snapshot.ts, snapshotTool.execute lacks focused valid- and rejected-input coverage, leaving delegated-result regressions undetected—add tests for both execution outcomes.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/integrations/eve/extension/tools/snapshot.ts">

<violation number="1" location="packages/integrations/eve/extension/tools/snapshot.ts:13">
P3: The new snapshot execution path has no focused test; the current suite only checks its metadata. Add valid-input and rejected-input tests that exercise `snapshotTool.execute` and verify the delegated result.

(Based on your team's feedback about adding unit tests for new behavior.)</violation>
</file>

<file name="scripts/release/consolidate-changelogs.ts">

<violation number="1" location="scripts/release/consolidate-changelogs.ts:30">
P2: Adding Eve to `packageChangelogs` changes the release consolidation set, but the release tests never exercise this package path or label. Add a focused test for an Eve-generated changelog so a future path or label regression cannot omit Eve release notes from the root changelog.

(Based on your team's feedback about tests for new behavior.) .</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Adding Eve to packageChangelogs changes the release consolidation set, but the release tests never exercise this package path or label. Add a focused test for an Eve-generated changelog so a future path or label regression cannot omit Eve release notes from the root changelog.

(Based on your team's feedback about tests for new behavior.) .

View Feedback

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At scripts/release/consolidate-changelogs.ts, line 30:

<comment>Adding Eve to `packageChangelogs` changes the release consolidation set, but the release tests never exercise this package path or label. Add a focused test for an Eve-generated changelog so a future path or label regression cannot omit Eve release notes from the root changelog.

(Based on your team's feedback about tests for new behavior.) .</comment>

<file context>
@@ -25,6 +25,10 @@ const packageChangelogs = [
   },
+  {
+    label: "Eve Extension",
+    path: path.join(repositoryRoot, "packages/integrations/eve/CHANGELOG.md"),
+  },
 ];
</file context>

[shrey150]

Addressed in 4016cf563: exported the consolidation source list and added a focused test for the exact Eve changelog path, label, and generated root heading.

[cubic-dev-ai]

P3: The new snapshot execution path has no focused test; the current suite only checks its metadata. Add valid-input and rejected-input tests that exercise snapshotTool.execute and verify the delegated result.

(Based on your team's feedback about adding unit tests for new behavior.)

View Feedback

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/integrations/eve/extension/tools/snapshot.ts, line 13:

<comment>The new snapshot execution path has no focused test; the current suite only checks its metadata. Add valid-input and rejected-input tests that exercise `snapshotTool.execute` and verify the delegated result.

(Based on your team's feedback about adding unit tests for new behavior.) </comment>

<file context>
@@ -0,0 +1,17 @@
+export default defineTool({
+  description: SNAPSHOT_TOOL_DESCRIPTION,
+  inputSchema: SNAPSHOT_INPUT_SCHEMA,
+  async execute(rawInput) {
+    const input = SnapshotInputSchema.parse(rawInput);
+    return stagehandSession.run(({ tools }) => tools.snapshot(input));
</file context>

[shrey150]

Addressed in 4016cf563: added execution coverage proving valid snapshot input delegates with the parsed options and invalid input rejects before the session is opened.